A score is a navigation instrument. It should compress prompt-level evidence into one traceable number, and every point on it should be explainable back to a specific prompt, platform, and observation. If a score can't be decomposed, it isn't measurement — it's marketing.
What you should leave with
- A score summarizes a sampled prompt set — it is not total market demand or a guaranteed rank.
- Recommendation coverage should carry the most weight; citations and mentions matter but are not the same signal.
- Confidence depends on prompt count, repeats, and platform coverage — a 20-prompt free audit is directional, not final.
- A low score is only useful when it points to a specific missing prompt family, source gap, or platform gap.
What should an AI Visibility Score measure?
An AI Visibility Score should summarize how often and how strongly a brand appears in AI-generated buyer answers, while keeping recommendations, mentions, citations, and factual accuracy separate underneath. It compresses a sampled prompt set into one navigable number, nothing more.
The score should be built from a defined prompt set run across chosen platforms, then rolled into a 0-100 figure. Every component feeding that number must remain visible and auditable, not buried inside a black-box calculation.
What it must never claim: total market demand, a universal rank across all users, or the probability any single searcher sees the brand. It is a sampled diagnostic, and treating it as a forecast misuses the instrument.
| Component | Example observation | Why it affects score |
|---|---|---|
| Recommendation | Brand named as the pick for a comparison prompt | Direct commercial signal |
| Citation | Brand's page linked as a source, no endorsement | Diagnostic, not a win |
| Mention | Brand named in a list, neutral tone | Weak visibility signal |
| Accuracy | Answer states wrong pricing or feature | Should reduce trust in score |
Evidence used in this section
What formula should the score use?
A defensible score weights direct recommendations highest, then adjusts for answer position, prompt intent, platform importance, citation support, factual accuracy, and repeat stability. No single input should dominate the number by accident.
Suggested default weights: recommendation coverage 40%, position/prominence 15%, prompt intent value 15%, competitor share-of-voice context 10%, citation/source support 10%, accuracy and stability 10%.
These are starting weights, not settled science, and every team should adjust them to its own commercial priorities.
- Freeze weights before measurement, not after seeing results.
- Document any weight change with a date and reason.
- See /measure-ai-share-of-voice for the share-of-voice component detail.
| Weight input | Default share |
|---|---|
| Recommendation coverage | 40% |
| Position/prominence | 15% |
| Prompt intent value | 15% |
| Competitor SOV context | 10% |
| Citation/source support | 10% |
| Accuracy and stability | 10% |
Evidence used in this section
Which observations should not be blended into one number?
Mentions, citations, recommendations, and negative appearances should be reported separately before they are summarized, because each means a different commercial outcome. Collapsing them early hides the exact problem a team needs to fix.
A citation-only result is diagnostic: it shows the brand's content was pulled in as a source, but it did not earn the answer's recommendation. Treating citation volume as a proxy for demand overstates what happened in that prompt run.
A negative mention — wrong claim, unfavorable comparison, outdated pricing — is not a visibility win even though the brand name appeared. Scoring engines must flag sentiment and correctness before folding an appearance into the total.
| Observation | Counts toward score? | Report separately? |
|---|---|---|
| Recommendation | Yes, full weight | Yes |
| Citation only | Partial weight | Yes |
| Neutral mention | Small weight | Yes |
| Negative mention | Reduces score | Yes |
How should confidence be shown?
Confidence should reflect prompt coverage, repeat count, platform coverage, entity-match certainty, citation availability, and human review quality. A score without a confidence label invites false precision and bad decisions.
Low confidence: small prompt set, single run, one platform, no human review. Medium confidence: broader prompt set, at least one repeat, multiple platforms, spot-checked entity matches. High confidence: repeated runs across cycles, full platform coverage, and reviewed accuracy against source pages.
A 20-prompt free audit should be read as directional — enough to spot obvious gaps, not enough to set quarterly targets. A baseline and retest, described at /ai-visibility-baseline-method, is what converts a directional read into a trackable trend.
- Low: 1 platform, 1 run, small prompt set.
- Medium: 2-3 platforms, 1+ repeat, spot-checked.
- High: multi-cycle repeats, full platform coverage, reviewed accuracy.
Evidence used in this section
What should a low score tell the team to do?
A low score is useful only when it points to a specific missing prompt family, competitor reason, source gap, factual error, or crawlability problem. Otherwise it is a number without a next step, and that is a wasted report.
Common low-score patterns worth separating: low recommendation coverage paired with high citation coverage often means content gets pulled in but never wins the pick — check comparison framing. High mentions with low direct recommendations usually signals a positioning gap, not a visibility gap.
Strong presence in ChatGPT but absence in Google AI features frequently traces to indexing or structured content differences between platforms, per Google's own guidance on AI features. A score drop caused by an expanded or edited prompt set is a measurement artifact, not proof the market moved.
- Low recommendations + high citations: check framing, not presence.
- High mentions + low recommendations: positioning gap.
- Strong in ChatGPT, absent in Google AI: check indexing/crawlability.
- Score drop from prompt-set edits: not real market movement.
What are the score's boundaries?
The score is a sampled diagnostic, not a public ranking factor, traffic forecast, or guarantee that the same user will see the same answer. Treat it as one signal among several, tracked over time, not as a verdict.
Structured data cannot be scored as a success on its own; schema only matters when it corresponds to content that is actually visible and cited in answers, consistent with Google's structured data policies. Markup without a matching answer is not a visibility gain.
Avoid comparing your score to invented or unsourced industry benchmarks. Any cross-company comparison should link to an actual, dated report, such as /reports/ai-visibility-benchmarks, rather than assert a universal average.
Method boundary: A score is not a ranking factor, forecast, or per-user guarantee. Schema-only changes should never be scored as a win without a matching visible answer improvement.
Evidence used in this section
Questions that change the decision
Frequently asked questions
What is a good AI Visibility Score?
There is no universal "good" number — a good score is one that's rising against your own baseline, driven by recommendation coverage, with medium-to-high confidence behind it. Compare to your history, not an invented industry average.
Should citations count toward AI visibility?
Partially. Citations are diagnostic evidence that content was surfaced as a source, but they should carry less weight than recommendations and always be reported separately so the difference stays visible.
Can I compare scores across platforms?
Only with caution. Google AI features and chat-style tools surface content differently, so a merged cross-platform score can mask which platform is actually driving or losing visibility.
Why did my AI Visibility Score change?
Common causes: prompt-set edits, platform answer variability, competitor content changes, or a genuine shift in recommendation coverage. See /why-ai-visibility-scores-change for how to separate measurement noise from real movement.
Is an AI Visibility Score the same as SEO visibility?
No. SEO visibility tracks rankings and organic traffic; an AI Visibility Score tracks appearance inside generated answers. They correlate sometimes but measure fundamentally different surfaces.
Primary sources and research
Platform documentation supports factual statements. Where we describe an audit method or prioritization rule, that is AnswerMentions' operating judgment and is labeled as such.
- [1]Google Search Central: AI features and your websiteGoogle explains how AI features can surface links and how crawlable, eligible web content remains part of the Search foundation.
- [2]arXiv: repeated measurement of AI search resultsThe paper supports repeated observations and careful treatment of answer variability instead of relying on one run.
- [3]Aggarwal et al.: Generative Engine OptimizationThe GEO research provides precedent for measuring visibility in generative engines, while not acting as a universal ranking recipe.
- [4]Google Search Central: structured data policiesGoogle states structured data should represent visible page content and follow its feature-specific guidelines.
- [5]NIST AI Risk Management FrameworkNIST frames AI measurement as a governed, monitored process with transparency, validity, and risk awareness.